1. Image gradients and edges Tuesday, September 1 st 2015 Devi Parikh Virginia Tech Disclaimer: Many slides have been borrowed from Kristen Grauman, who may have borrowed some of them from others. Any time a slide did not already have a credit on it, I have credited it to Kristen. So there is a chance some of these credits are inaccurate.

2. Announcements Questions? PS0 was due last night –5 late PS1 is out –Due September 16 th (Wednesday) 11:55 pm 2 Slide credit: Devi Parikh

3. Topics overview Features & filters –Filters Grouping & fitting Multiple views and motion Recognition Video processing 3 Slide credit: Kristen Grauman

4. Topics overview Features & filters –Filters –Gradients –Edges Grouping & fitting Multiple views and motion Recognition Video processing 4 Slide credit: Kristen Grauman

5. Last time Image formation Various models for image “noise” Linear filters and convolution useful for –Image smoothing, removing noise Box filter Gaussian filter Impact of scale / width of smoothing filter Separable filters more efficient Median filter: a non-linear filter, edge-preserving 5 Slide credit: Adapted by Devi Parikh from Kristen Grauman

6. f*g=? original image hfiltered Filter f = 1/9 x [ 1 1 1 1 1 1 1 1 1] Review 6 Slide credit: Kristen Grauman

7. f*g=? Filter f = 1/9 x [ 1 1 1 1 1 1 1 1 1] T original image hfiltered Review 7 Slide credit: Kristen Grauman

8. Review How do you sharpen an image? 8 Slide credit: Devi Parikh

9. Review Median filter f: Is f(a+b) = f(a)+f(b)? Example: a = [10 20 30 40 50] b = [55 20 30 40 50] Is f linear? 9 Slide credit: Devi Parikh

10. Recall: image filtering Compute a function of the local neighborhood at each pixel in the image –Function specified by a “filter” or mask saying how to combine values from neighbors. Uses of filtering: –Enhance an image (denoise, resize, etc) –Extract information (texture, edges, etc) –Detect patterns (template matching) 10 Slide credit: Kristen Grauman, Adapted from Derek Hoiem

11. Recall: image filtering Compute a function of the local neighborhood at each pixel in the image –Function specified by a “filter” or mask saying how to combine values from neighbors. Uses of filtering: –Enhance an image (denoise, resize, etc) –Extract information (texture, edges, etc) –Detect patterns (template matching) 11 Slide credit: Kristen Grauman, Adapted from Derek Hoiem

12. Edge detection Goal: map image from 2d array of pixels to a set of curves or line segments or contours. Why? Main idea: look for strong gradients, post-process Figure from J. Shotton et al., PAMI 2007 12 Slide credit: Kristen Grauman

13. What causes an edge? Depth discontinuity: object boundary Change in surface orientation: shape Cast shadows Reflectance change: appearance information, texture 13 Slide credit: Kristen Grauman

14. Edges/gradients and invariance 14 Slide credit: Kristen Grauman

15. Derivatives and edges image intensity function (along horizontal scanline) first derivative edges correspond to extrema of derivative An edge is a place of rapid change in the image intensity function. 15 Slide credit: Svetlana Lazebnik

16. Derivatives with convolution For 2D function, f(x,y), the partial derivative is: For discrete data, we can approximate using finite differences: To implement above as convolution, what would be the associated filter? 16 Slide credit: Kristen Grauman

17. Partial derivatives of an image Which shows changes with respect to x? -1 1 1 -1 or ? -1 1 (showing filters for correlation) 17 Slide credit: Kristen Grauman

18. Assorted finite difference filters >> My = fspecial(‘sobel’); >> outim = imfilter(double(im), My); >> imagesc(outim); >> colormap gray; 18 Slide credit: Kristen Grauman

19. Image gradient The gradient of an image: The gradient points in the direction of most rapid change in intensity The gradient direction (orientation of edge normal) is given by: The edge strength is given by the gradient magnitude 19 Slide credit: Steve Seitz

20. Effects of noise Consider a single row or column of the image Plotting intensity as a function of position gives a signal Where is the edge? 20 Slide credit: Steve Seitz

21. Where is the edge? Solution: smooth first Look for peaks in 21 Slide credit: Kristen Grauman

22. Derivative theorem of convolution Differentiation property of convolution. 22 Slide credit: Steve Seitz

23. 0.0030 0.0133 0.0219 0.0133 0.0030 0.0133 0.0596 0.0983 0.0596 0.0133 0.0219 0.0983 0.1621 0.0983 0.0219 0.0133 0.0596 0.0983 0.0596 0.0133 0.0030 0.0133 0.0219 0.0133 0.0030 Derivative of Gaussian filters 23 Slide credit: Kristen Grauman

24. Derivative of Gaussian filters x-direction y-direction 24 Slide credit: Svetlana Lazebnik

25. Derivative theorem of convolution 25 Slide credit: Steve Seitz

26. Laplacian of Gaussian Consider Laplacian of Gaussian operator Where is the edge?Zero-crossings of bottom graph 26 Slide credit: Steve Seitz

27. 2D edge detection filters is the Laplacian operator: Laplacian of Gaussian Gaussianderivative of Gaussian 27 Slide credit: Steve Seitz

28. Smoothing with a Gaussian Recall: parameter σ is the “scale” / “width” / “spread” of the Gaussian kernel, and controls the amount of smoothing. … 28 Slide credit: Kristen Grauman

29. Effect of σ on derivatives The apparent structures differ depending on Gaussian’s scale parameter. Larger values: larger scale edges detected Smaller values: finer features detected σ = 1 pixel σ = 3 pixels 29 Slide credit: Kristen Grauman

30. So, what scale to choose? It depends what we’re looking for. 30 Slide credit: Kristen Grauman

31. Mask properties Smoothing –Values positive –Sum to 1  constant regions same as input –Amount of smoothing proportional to mask size –Remove “high-frequency” components; “low-pass” filter Derivatives –___________ signs used to get high response in regions of high contrast –Sum to ___  no response in constant regions –High absolute value at points of high contrast 31 Slide credit: Kristen Grauman

32. Seam carving: main idea [Shai & Avidan, SIGGRAPH 2007] 32 Slide credit: Kristen Grauman

33. Content-aware resizing Traditional resizing Seam carving: main idea [Shai & Avidan, SIGGRAPH 2007] 33 Slide credit: Kristen Grauman

34. Seam carving: main idea 34 Slide credit: Devi Parikh

35. Content-aware resizing Seam carving: main idea Intuition: Preserve the most “interesting” content  Prefer to remove pixels with low gradient energy To reduce or increase size in one dimension, remove irregularly shaped “seams”  Optimal solution via dynamic programming. 35 Slide credit: Kristen Grauman

36. Want to remove seams where they won’t be very noticeable: –Measure “energy” as gradient magnitude Choose seam based on minimum total energy path across image, subject to 8-connectedness. Seam carving: main idea 36 Slide credit: Kristen Grauman

37. 37 Let a vertical seam s consist of h positions that form an 8-connected path. Let the cost of a seam be: Optimal seam minimizes this cost: Compute it efficiently with dynamic programming. Seam carving: algorithm s1s1 s2s2 s3s3 s4s4 s5s5 Slide credit: Kristen Grauman

38. How to identify the minimum cost seam? How many possible seams are there? First, consider a greedy approach: Energy matrix (gradient magnitude) 38 Slide credit: Adapted by Devi Parikh from Kristen Grauman

39. Seam carving: algorithm 39 Slide credit: Devi Parikh If I tell you the minimum cost over all seams starting at row 1 and ending at this point For every pixel in this row: M(i,j) What is the minimum cost over all seams starting at row 1 and ending here?

40. row i-1 Seam carving: algorithm Compute the cumulative minimum energy for all possible connected seams at each entry (i,j): Then, min value in last row of M indicates end of the minimal connected vertical seam. Backtrack up from there, selecting min of 3 above in M. j-1 j row i M matrix: cumulative min energy (for vertical seams) Energy matrix (gradient magnitude) j j+1 40 Slide credit: Kristen Grauman

41. Example Energy matrix (gradient magnitude) M matrix (for vertical seams) 41 Slide credit: Kristen Grauman

42. Example Energy matrix (gradient magnitude) M matrix (for vertical seams) 42 Slide credit: Kristen Grauman

43. Real image example Original Image Energy Map Blue = low energy Red = high energy 43 Slide credit: Kristen Grauman

44. Real image example 44 Slide credit: Kristen Grauman

45. Other notes on seam carving Analogous procedure for horizontal seams Can also insert seams to increase size of image in either dimension –Duplicate optimal seam, averaged with neighbors Other energy functions may be plugged in –E.g., color-based, interactive,… Can use combination of vertical and horizontal seams 45 Slide credit: Kristen Grauman

46. Results from Eunho Yang Example results from students at UT Austin 46 Slide credit: Kristen Grauman

47. Results from Suyog Jain 47 Slide credit: Kristen Grauman

48. Seam carving result Original image Results from Martin Becker Conventional resize 48 Slide credit: Kristen Grauman

49. Seam carving result Conventional resize Original image Results from Martin Becker 49 Slide credit: Kristen Grauman

50. Results from Jay Hennig Original image (599 by 799) Conventional resize (399 by 599) Seam carving (399 by 599) 50 Slide credit: Kristen Grauman

51. Results from Donghyuk Shin Removal of a marked object 51 Slide credit: Kristen Grauman

52. Removal of a marked object Results from Eunho Yang 52 Slide credit: Kristen Grauman

53. “Failure cases” with seam carving By Donghyuk Shin 53 Slide credit: Kristen Grauman

54. “Failure cases” with seam carving By Suyog Jain 54 Slide credit: Kristen Grauman

55. Topics overview Features & filters – Filters – Gradients – Edges Grouping & fitting Multiple views and motion Recognition Video processing 55 Slide credit: Kristen Grauman

56. Questions? See you Thursday!